Re: [PATCH] add i386 idle notifier (take 3)

[Stephane Eranian <eranian@hpl.hp.com>]

Adrian,

On Sat, Dec 23, 2006 at 12:40:15PM +0100, Adrian Bunk wrote:
> > 
> > If you look at the perfmon-new-base patch, you'll see a base.diff patch which
> > includes this one. I am slowly getting rid of this requirement by pushing
> > those "infrastructure patches" to mainline so that the perfmon patch gets
> > smaller over time. Submitting smaller patches makes it easier for maintainers
> > to integrate.
> 
> No, the preferred way is to start with getting both the infrastructure 
> and the users into -mm.
> 
> Adding infrastructure without users doesn't fit into the kernel 
> development model.
> 

I am hearing conflicting opinions on this one.

Perfmon is a fairly big patch. It is hard to take it as one. I have tried to
split it up in smaller, more manageable pieces as requested by top-level
maintainers. This process implies that I supply small patches which may not
necessarily have users just yet.

> The unused x86-64 idle notifiers are now bloating the kernel since 
> nearly one year.
> 
> > > And why does it bloat the kernel with EXPORT_SYMBOL's although even your 
> > > perfmon-new-base-061204 doesn't seem to add any modular user?
> > 
> Where does the perfmon code use the EXPORT_SYMBOL's?

The perfmon patch includes several kernel modules which make use of
the exported entry points. The following symbols are exported:

pfm_pmu_register/pfm_pmu_unregister:
	* PMU description module registration.
	* Used to describe PMU model.
	* Used by perfmon_p4.c, perfmon_core.c, perfmon_mckinley.c, and others

pfm_fmt_register/pfm_fmt_unregister:
	* Sampling format module registration
	* Used by perfmon_dfl_smpl.c, perfmon_pebs_smpl.c

pfm_interrupt_handler:
	* PMU interrupt handler
	* Used by MIPS-specific perfmon code

pfm_pmu_conf/pfm_controls:
	* global state/control variable

All exported symbols are currently used. Why are you saying this adds bloat?

-- 
-Stephane